Artificial Intelligence Nanodegree

Convolutional Neural Networks

Project: Write an Algorithm for a Dog Identification App


In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!

Note: Once you have completed all of the code implementations, you need to finalize your work by exporting the iPython Notebook as an HTML document. Before exporting the notebook to html, all of the code cells need to have been run so that reviewers can see the final implementation and output. You can then export the notebook by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.

In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.

The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this IPython notebook.


Why We're Here

In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).

Sample Dog Output

In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!

The Road Ahead

We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.

  • Step 0: Import Datasets
  • Step 1: Detect Humans
  • Step 2: Detect Dogs
  • Step 3: Create a CNN to Classify Dog Breeds (from Scratch)
  • Step 4: Use a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 6: Write your Algorithm
  • Step 7: Test Your Algorithm

Step 0: Import Datasets

Import Dog Dataset

In the code cell below, we import a dataset of dog images. We populate a few variables through the use of the load_files function from the scikit-learn library:

  • train_files, valid_files, test_files - numpy arrays containing file paths to images
  • train_targets, valid_targets, test_targets - numpy arrays containing onehot-encoded classification labels
  • dog_names - list of string-valued dog breed names for translating labels
In [2]:
from sklearn.datasets import load_files       
from keras.utils import np_utils
import numpy as np
from glob import glob

# define function to load train, test, and validation datasets
def load_dataset(path):
    data = load_files(path)
    dog_files = np.array(data['filenames'])
    dog_targets = np_utils.to_categorical(np.array(data['target']), 133)
    return dog_files, dog_targets

# load train, test, and validation datasets
train_files, train_targets = load_dataset('dogImages/train')
valid_files, valid_targets = load_dataset('dogImages/valid')
test_files, test_targets = load_dataset('dogImages/test')

# load list of dog names
dog_names = [item[20:-1] for item in sorted(glob("dogImages/train/*/"))]

# print statistics about the dataset
print('There are %d total dog categories.' % len(dog_names))
print('There are %s total dog images.\n' % len(np.hstack([train_files, valid_files, test_files])))
print('There are %d training dog images.' % len(train_files))
print('There are %d validation dog images.' % len(valid_files))
print('There are %d test dog images.'% len(test_files))
There are 133 total dog categories.
There are 8351 total dog images.

There are 6680 training dog images.
There are 835 validation dog images.
There are 836 test dog images.

Import Human Dataset

In the code cell below, we import a dataset of human images, where the file paths are stored in the numpy array human_files.

In [3]:
import random
random.seed(8675309)

# load filenames in shuffled human dataset
human_files = np.array(glob("lfw/*/*"))
random.shuffle(human_files)

# print statistics about the dataset
print('There are %d total human images.' % len(human_files))
There are 13233 total human images.

Step 1: Detect Humans

We use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images. OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory.

In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.

In [4]:
import cv2                
import matplotlib.pyplot as plt                        
%matplotlib inline                               

# extract pre-trained face detector
face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt.xml')

# load color (BGR) image
img = cv2.imread(human_files[3])
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# find faces in image
faces = face_cascade.detectMultiScale(gray)

# print number of faces detected in the image
print('Number of faces detected:', len(faces))

# get bounding box for each detected face
for (x,y,w,h) in faces:
    # add bounding box to color image
    cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
    
# convert BGR image to RGB for plotting
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# display the image, along with bounding box
plt.imshow(cv_rgb)
plt.show()
Number of faces detected: 1

Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.

In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.

Write a Human Face Detector

We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.

In [5]:
# returns "True" if face is detected in image stored at img_path
def face_detector(img_path):
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray)
    return len(faces) > 0

(IMPLEMENTATION) Assess the Human Face Detector

Question 1: Use the code cell below to test the performance of the face_detector function.

  • What percentage of the first 100 images in human_files have a detected human face?
  • What percentage of the first 100 images in dog_files have a detected human face?

Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.

Answer:

In [7]:
human_files_short = human_files[:100]
dog_files_short = train_files[:100]
# Do NOT modify the code above this line.

def percentage_predicted(img_file, detector):
    return 100.0*([detector(i) for i in img_file].count(True))/len(img_file)

print('{:.2f}% of the first 100 images in human_files have a detected human face.'.format(
    percentage_predicted(human_files_short, face_detector)))
print('{:.2f}% of the first 100 images in dog_files have a detected human face.'.format(
    percentage_predicted(dog_files_short, face_detector)))
99.00% of the first 100 images in human_files have a detected human face.
11.00% of the first 100 images in dog_files have a detected human face.

Question 2: This algorithmic choice necessitates that we communicate to the user that we accept human images only when they provide a clear view of a face (otherwise, we risk having unneccessarily frustrated users!). In your opinion, is this a reasonable expectation to pose on the user? If not, can you think of a way to detect humans in images that does not necessitate an image with a clearly presented face?

Answer:

In this case, we are not dealing with identity identification, just simple human face detection. So IMHO, it's not necessary to ask users to provide a clear view of a face. In order to detect humans in images without clearly presented face, we might use CNN with Harr-features and/or HOG-features.

We suggest the face detector from OpenCV as a potential way to detect human images in your algorithm, but you are free to explore other approaches, especially approaches that make use of deep learning :). Please use the code cell below to design and test your own face detection algorithm. If you decide to pursue this optional task, report performance on each of the datasets.

In [ ]:
## (Optional) TODO: Report the performance of another  
## face detection algorithm on the LFW dataset
### Feel free to use as many code cells as needed.
In [8]:
!python3 -m pip install dlib
Collecting dlib
  Downloading https://files.pythonhosted.org/packages/f2/02/dbffb7023494e1e39981ca65010b27501e844a0c8e79e1a3034ad8bf3734/dlib-19.12.0.tar.gz (3.3MB)
    100% |████████████████████████████████| 3.3MB 211kB/s eta 0:00:01
Building wheels for collected packages: dlib
  Running setup.py bdist_wheel for dlib ... done
  Stored in directory: /root/.cache/pip/wheels/10/c7/a9/dfd512f1a3d9e31188919b5a8cd79cea1babf09b79761a849a
Successfully built dlib
Installing collected packages: dlib
Successfully installed dlib-19.12.0
You are using pip version 9.0.1, however version 10.0.1 is available.
You should consider upgrading via the 'pip install --upgrade pip' command.
In [9]:
import dlib

def face_detector2(img_path):
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    detector = dlib.get_frontal_face_detector()
    faces1 = detector(gray, 0)
    faces2 = face_cascade.detectMultiScale(gray)
    if len(faces1) > 0 and len(faces2) > 0:
        return True
    else:
        return False

We use dlib(HOG-features) and cv2.CascadeClassifier(Harr-features) to double-check whether there is/are human face(s) in the images. As shown below, in the first 200 images in human_files, new face_detector2 can regonize 98% of faces correctly, just like the original face_detector. And in the first 200 images in train_files, new face_detector2 found less percentage of human faces in the images.

In [10]:
print('Using the original face_detector:')
print('{:.2f}% of the first 200 images in human_files have a detected human face.'.format(
    percentage_predicted(human_files[:200], face_detector)))

print('\nUsing the new face_detector2:')
print('{:.2f}% of the first 200 images in human_files have a detected human face.'.format(
    percentage_predicted(human_files[:200], face_detector2)))
Using the original face_detector:
98.00% of the first 200 images in human_files have a detected human face.

Using the new face_detector2:
98.00% of the first 200 images in human_files have a detected human face.
In [11]:
print('Using the original face_detector:')
print('{:.2f}% of the first 200 images in train_files have a detected human face.'.format(
    percentage_predicted(train_files[:200], face_detector)))

print('\nUsing the new face_detector2:')
print('{:.2f}% of the first 200 images in train_files have a detected human face.'.format(
    percentage_predicted(train_files[:200], face_detector2)))
Using the original face_detector:
13.00% of the first 200 images in train_files have a detected human face.

Using the new face_detector2:
2.00% of the first 200 images in train_files have a detected human face.

First, we want to know what kind of human faces in the images cannot be detected. As shown below, even provided a clear view of a face, there still some faces could not be recognized by face_detector2.

In [12]:
from IPython.core.display import display, HTML
display(HTML("<style>.container {width:90% !important; }</style>"))

from scipy import misc, ndimage

def plot_face_detect(img_files, detector, faces=True):
    k = 0
    image_path = list()
    for i in img_files:
        if detector(i) == faces:
            image_path += [i]
    fig = plt.figure(figsize=(25,25))
    for j in image_path:
        fig.add_subplot(6, 6, k+1)
        plt.imshow(np.expand_dims(ndimage.imread(j), 0)[0])
        plt.title(j.split('/')[-1])
        k += 1
    plt.show()  
In [13]:
plot_face_detect(human_files[:200], face_detector2, faces=False)

Second, as we can see below, in the first 200 images in train_files, there are 26 images detected human faces in the image by face_detector, but only 6 images have human faces. On the other hand, new face_detector2 found 4 images have human faces, but just one image actually has no human face in it. Although face_detector2 cannot detect the other 3 images(at least) that has human face, it still better than the original face_detector in some applications.

In [14]:
plot_face_detect(train_files[:200], face_detector, faces=True)
In [15]:
plot_face_detect(train_files[:200], face_detector2, faces=True)

Step 2: Detect Dogs

In this section, we use a pre-trained ResNet-50 model to detect dogs in images. Our first line of code downloads the ResNet-50 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories. Given an image, this pre-trained ResNet-50 model returns a prediction (derived from the available categories in ImageNet) for the object that is contained in the image.

In [16]:
from keras.applications.resnet50 import ResNet50

# define ResNet50 model
ResNet50_model = ResNet50(weights='imagenet')
Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.2/resnet50_weights_tf_dim_ordering_tf_kernels.h5
102858752/102853048 [==============================] - 1s 0us/step: 

Pre-process the Data

When using TensorFlow as backend, Keras CNNs require a 4D array (which we'll also refer to as a 4D tensor) as input, with shape

$$ (\text{nb_samples}, \text{rows}, \text{columns}, \text{channels}), $$

where nb_samples corresponds to the total number of images (or samples), and rows, columns, and channels correspond to the number of rows, columns, and channels for each image, respectively.

The path_to_tensor function below takes a string-valued file path to a color image as input and returns a 4D tensor suitable for supplying to a Keras CNN. The function first loads the image and resizes it to a square image that is $224 \times 224$ pixels. Next, the image is converted to an array, which is then resized to a 4D tensor. In this case, since we are working with color images, each image has three channels. Likewise, since we are processing a single image (or sample), the returned tensor will always have shape

$$ (1, 224, 224, 3). $$

The paths_to_tensor function takes a numpy array of string-valued image paths as input and returns a 4D tensor with shape

$$ (\text{nb_samples}, 224, 224, 3). $$

Here, nb_samples is the number of samples, or number of images, in the supplied array of image paths. It is best to think of nb_samples as the number of 3D tensors (where each 3D tensor corresponds to a different image) in your dataset!

In [17]:
from keras.preprocessing import image                  
from tqdm import tqdm

def path_to_tensor(img_path):
    # loads RGB image as PIL.Image.Image type
    img = image.load_img(img_path, target_size=(224, 224))
    # convert PIL.Image.Image type to 3D tensor with shape (224, 224, 3)
    x = image.img_to_array(img)
    # convert 3D tensor to 4D tensor with shape (1, 224, 224, 3) and return 4D tensor
    return np.expand_dims(x, axis=0)

def paths_to_tensor(img_paths):
    list_of_tensors = [path_to_tensor(img_path) for img_path in tqdm(img_paths)]
    return np.vstack(list_of_tensors)

Making Predictions with ResNet-50

Getting the 4D tensor ready for ResNet-50, and for any other pre-trained model in Keras, requires some additional processing. First, the RGB image is converted to BGR by reordering the channels. All pre-trained models have the additional normalization step that the mean pixel (expressed in RGB as $[103.939, 116.779, 123.68]$ and calculated from all pixels in all images in ImageNet) must be subtracted from every pixel in each image. This is implemented in the imported function preprocess_input. If you're curious, you can check the code for preprocess_input here.

Now that we have a way to format our image for supplying to ResNet-50, we are now ready to use the model to extract the predictions. This is accomplished with the predict method, which returns an array whose $i$-th entry is the model's predicted probability that the image belongs to the $i$-th ImageNet category. This is implemented in the ResNet50_predict_labels function below.

By taking the argmax of the predicted probability vector, we obtain an integer corresponding to the model's predicted object class, which we can identify with an object category through the use of this dictionary.

In [18]:
from keras.applications.resnet50 import preprocess_input, decode_predictions

def ResNet50_predict_labels(img_path):
    # returns prediction vector for image located at img_path
    img = preprocess_input(path_to_tensor(img_path))
    return np.argmax(ResNet50_model.predict(img))

Write a Dog Detector

While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'. Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained ResNet-50 model, we need only check if the ResNet50_predict_labels function above returns a value between 151 and 268 (inclusive).

We use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).

In [19]:
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
    prediction = ResNet50_predict_labels(img_path)
    return ((prediction <= 268) & (prediction >= 151)) 

(IMPLEMENTATION) Assess the Dog Detector

Question 3: Use the code cell below to test the performance of your dog_detector function.

  • What percentage of the images in human_files_short have a detected dog?
  • What percentage of the images in dog_files_short have a detected dog?

Answer:

In [20]:
### Test the performance of the dog_detector function
### on the images in human_files_short and dog_files_short.

print('{:.2f}% of the first 100 images in human_files have a detected dog.'.format(
    percentage_predicted(human_files_short, dog_detector)))
print('{:.2f}% of the first 100 images in dog_files have a detected dog.'.format(
    percentage_predicted(dog_files_short, dog_detector)))
1.00% of the first 100 images in human_files have a detected dog.
100.00% of the first 100 images in dog_files have a detected dog.

Step 3: Create a CNN to Classify Dog Breeds (from Scratch)

Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 1%. In Step 5 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.

Be careful with adding too many trainable layers! More parameters means longer training, which means you are more likely to need a GPU to accelerate the training process. Thankfully, Keras provides a handy estimate of the time that each epoch is likely to take; you can extrapolate this estimate to figure out how long it will take for your algorithm to train.

We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have great difficulty in distinguishing between a Brittany and a Welsh Springer Spaniel.

Brittany Welsh Springer Spaniel

It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).

Curly-Coated Retriever American Water Spaniel

Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.

Yellow Labrador Chocolate Labrador Black Labrador

We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.

Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!

Pre-process the Data

We rescale the images by dividing every pixel in every image by 255.

In [21]:
from PIL import ImageFile                            
ImageFile.LOAD_TRUNCATED_IMAGES = True                 

# pre-process the data for Keras
train_tensors = paths_to_tensor(train_files).astype('float32')/255
valid_tensors = paths_to_tensor(valid_files).astype('float32')/255
test_tensors = paths_to_tensor(test_files).astype('float32')/255
100%|██████████| 6680/6680 [01:12<00:00, 92.26it/s] 
100%|██████████| 835/835 [00:07<00:00, 106.25it/s]
100%|██████████| 836/836 [00:07<00:00, 106.37it/s]

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    model.summary()

We have imported some Python modules to get you started, but feel free to import as many modules as you need. If you end up getting stuck, here's a hint that specifies a model that trains relatively fast on CPU and attains >1% test accuracy in 5 epochs:

Sample CNN

Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. If you chose to use the hinted architecture above, describe why you think that CNN architecture should work well for the image classification task.

Answer:

  1. Increase the number of filters(32->64->128->256) and decrease the kernel size(5->4->3->2) layer-by-layer, to extract more and more detail features.
  2. Use a new activation function(SWISH), which just proposed last year, to avoid 'dying ReLU' and vanishing gradient.
  3. Use batch normalizaion layer to normalize the activations of the previous layer at each batch(that maintains the activation mean and standard deviation close to 0 and 1 respectively). Due to the help of batch normalization, we can increase the learning rate to 2e-3(with Adam optimizer).
  4. In the end, we use GAP layer(Global Average Pooling) to generate a feature map for each corresponding category and feed directly into the softmax layer. With less parameters it avoids overfitting and speed-up the training time.

[ref-1] Searching for Activation Functions(Oct 2017)
[ref-2] Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift(Feb 2015)
[ref-3] Network In Network(Dec 2013)

In [22]:
from keras import backend as K
from keras.utils.generic_utils import get_custom_objects
def swish(x):
    return (K.sigmoid(x) * x)
get_custom_objects().update({'swish': swish})
In [23]:
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
from keras.layers import Dropout, Flatten, Dense, BatchNormalization
from keras.models import Sequential

model = Sequential()

### Define your architecture.
model.add(Conv2D(32, kernel_size=(5), activation='swish', input_shape=(224, 224, 3)))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(BatchNormalization(momentum=0.75, epsilon=0.001))

model.add(Conv2D(64, kernel_size=(4), activation='swish'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(BatchNormalization(momentum=0.75, epsilon=0.001))

model.add(Conv2D(128, kernel_size=(3), activation='swish'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(BatchNormalization(momentum=0.75, epsilon=0.001))

model.add(Conv2D(256, kernel_size=(2), activation='swish'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(BatchNormalization(momentum=0.75, epsilon=0.001))
model.add(GlobalAveragePooling2D())
model.add(Dense(133, activation='softmax'))

model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 220, 220, 32)      2432      
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 110, 110, 32)      0         
_________________________________________________________________
batch_normalization_1 (Batch (None, 110, 110, 32)      128       
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 107, 107, 64)      32832     
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 53, 53, 64)        0         
_________________________________________________________________
batch_normalization_2 (Batch (None, 53, 53, 64)        256       
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 51, 51, 128)       73856     
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 25, 25, 128)       0         
_________________________________________________________________
batch_normalization_3 (Batch (None, 25, 25, 128)       512       
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 24, 24, 256)       131328    
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 12, 12, 256)       0         
_________________________________________________________________
batch_normalization_4 (Batch (None, 12, 12, 256)       1024      
_________________________________________________________________
global_average_pooling2d_1 ( (None, 256)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 133)               34181     
=================================================================
Total params: 276,549
Trainable params: 275,589
Non-trainable params: 960
_________________________________________________________________

Compile the Model

In [24]:
from keras.optimizers import Adam
model.compile(optimizer=Adam(lr=2e-3), loss='categorical_crossentropy', metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [25]:
from keras.callbacks import ModelCheckpoint  

### specify the number of epochs that you would like to use to train the model.

epochs = 20

### Do NOT modify the code below this line.

checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.from_scratch_withdropout.hdf5', 
                               verbose=1, save_best_only=True)

hist = model.fit(train_tensors, train_targets, 
          validation_data=(valid_tensors, valid_targets),
          epochs=epochs, batch_size=20, callbacks=[checkpointer], verbose=1)
Train on 6680 samples, validate on 835 samples
Epoch 1/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.7182 - acc: 0.0302Epoch 00001: val_loss improved from inf to 4.62022, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 56s 8ms/step - loss: 4.7168 - acc: 0.0302 - val_loss: 4.6202 - val_acc: 0.0395
Epoch 2/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.3501 - acc: 0.0616Epoch 00002: val_loss improved from 4.62022 to 4.60979, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 54s 8ms/step - loss: 4.3493 - acc: 0.0615 - val_loss: 4.6098 - val_acc: 0.0623
Epoch 3/20
6660/6680 [============================>.] - ETA: 0s - loss: 4.0167 - acc: 0.0985Epoch 00003: val_loss improved from 4.60979 to 4.10106, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 54s 8ms/step - loss: 4.0163 - acc: 0.0987 - val_loss: 4.1011 - val_acc: 0.0934
Epoch 4/20
6660/6680 [============================>.] - ETA: 0s - loss: 3.7377 - acc: 0.1350Epoch 00004: val_loss improved from 4.10106 to 4.04444, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 54s 8ms/step - loss: 3.7363 - acc: 0.1353 - val_loss: 4.0444 - val_acc: 0.1030
Epoch 5/20
6660/6680 [============================>.] - ETA: 0s - loss: 3.4397 - acc: 0.1872Epoch 00005: val_loss did not improve
6680/6680 [==============================] - 54s 8ms/step - loss: 3.4407 - acc: 0.1868 - val_loss: 4.0751 - val_acc: 0.1257
Epoch 6/20
6660/6680 [============================>.] - ETA: 0s - loss: 3.1224 - acc: 0.2446Epoch 00006: val_loss improved from 4.04444 to 3.74699, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 54s 8ms/step - loss: 3.1228 - acc: 0.2445 - val_loss: 3.7470 - val_acc: 0.1317
Epoch 7/20
6660/6680 [============================>.] - ETA: 0s - loss: 2.8642 - acc: 0.2952Epoch 00007: val_loss improved from 3.74699 to 3.47549, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 54s 8ms/step - loss: 2.8640 - acc: 0.2952 - val_loss: 3.4755 - val_acc: 0.2120
Epoch 8/20
6660/6680 [============================>.] - ETA: 0s - loss: 2.5683 - acc: 0.3491Epoch 00008: val_loss improved from 3.47549 to 3.19903, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 54s 8ms/step - loss: 2.5679 - acc: 0.3490 - val_loss: 3.1990 - val_acc: 0.2251
Epoch 9/20
6660/6680 [============================>.] - ETA: 0s - loss: 2.3257 - acc: 0.4111Epoch 00009: val_loss did not improve
6680/6680 [==============================] - 54s 8ms/step - loss: 2.3251 - acc: 0.4112 - val_loss: 3.2559 - val_acc: 0.2443
Epoch 10/20
6660/6680 [============================>.] - ETA: 0s - loss: 2.0414 - acc: 0.4769Epoch 00010: val_loss improved from 3.19903 to 3.07617, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 54s 8ms/step - loss: 2.0415 - acc: 0.4769 - val_loss: 3.0762 - val_acc: 0.2826
Epoch 11/20
6660/6680 [============================>.] - ETA: 0s - loss: 1.7878 - acc: 0.5276Epoch 00011: val_loss improved from 3.07617 to 2.85650, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 54s 8ms/step - loss: 1.7892 - acc: 0.5272 - val_loss: 2.8565 - val_acc: 0.2910
Epoch 12/20
6660/6680 [============================>.] - ETA: 0s - loss: 1.5404 - acc: 0.5977Epoch 00012: val_loss improved from 2.85650 to 2.62920, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 54s 8ms/step - loss: 1.5407 - acc: 0.5976 - val_loss: 2.6292 - val_acc: 0.3401
Epoch 13/20
6660/6680 [============================>.] - ETA: 0s - loss: 1.3273 - acc: 0.6530Epoch 00013: val_loss did not improve
6680/6680 [==============================] - 54s 8ms/step - loss: 1.3276 - acc: 0.6527 - val_loss: 2.6588 - val_acc: 0.3557
Epoch 14/20
6660/6680 [============================>.] - ETA: 0s - loss: 1.0950 - acc: 0.7266Epoch 00014: val_loss did not improve
6680/6680 [==============================] - 54s 8ms/step - loss: 1.0956 - acc: 0.7265 - val_loss: 2.6460 - val_acc: 0.3497
Epoch 15/20
6660/6680 [============================>.] - ETA: 0s - loss: 0.8973 - acc: 0.7730Epoch 00015: val_loss did not improve
6680/6680 [==============================] - 54s 8ms/step - loss: 0.8959 - acc: 0.7735 - val_loss: 2.6906 - val_acc: 0.3413
Epoch 16/20
6660/6680 [============================>.] - ETA: 0s - loss: 0.7320 - acc: 0.8261Epoch 00016: val_loss did not improve
6680/6680 [==============================] - 54s 8ms/step - loss: 0.7323 - acc: 0.8259 - val_loss: 2.8429 - val_acc: 0.3293
Epoch 17/20
6660/6680 [============================>.] - ETA: 0s - loss: 0.5575 - acc: 0.8746Epoch 00017: val_loss did not improve
6680/6680 [==============================] - 54s 8ms/step - loss: 0.5576 - acc: 0.8747 - val_loss: 2.6891 - val_acc: 0.3796
Epoch 18/20
6660/6680 [============================>.] - ETA: 0s - loss: 0.4437 - acc: 0.9069Epoch 00018: val_loss improved from 2.62920 to 2.60365, saving model to saved_models/weights.best.from_scratch_withdropout.hdf5
6680/6680 [==============================] - 54s 8ms/step - loss: 0.4438 - acc: 0.9067 - val_loss: 2.6037 - val_acc: 0.3844
Epoch 19/20
6660/6680 [============================>.] - ETA: 0s - loss: 0.3454 - acc: 0.9326Epoch 00019: val_loss did not improve
6680/6680 [==============================] - 54s 8ms/step - loss: 0.3467 - acc: 0.9325 - val_loss: 2.8585 - val_acc: 0.3605
Epoch 20/20
6660/6680 [============================>.] - ETA: 0s - loss: 0.2850 - acc: 0.9485Epoch 00020: val_loss did not improve
6680/6680 [==============================] - 54s 8ms/step - loss: 0.2851 - acc: 0.9487 - val_loss: 2.7541 - val_acc: 0.3820

Load the Model with the Best Validation Loss

In [26]:
model.load_weights('saved_models/weights.best.from_scratch_withdropout.hdf5')

Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 1%.

In [27]:
# get index of predicted dog breed for each image in test set
dog_breed_predictions = [np.argmax(model.predict(np.expand_dims(tensor, axis=0))) for tensor in test_tensors]

# report test accuracy
test_accuracy = 100*np.sum(np.array(dog_breed_predictions)==np.argmax(test_targets, axis=1))/len(dog_breed_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 40.7895%
In [28]:
import matplotlib.pyplot as plt
def plot_model(model_details):

    # Create sub-plots
    fig, axs = plt.subplots(1,2,figsize=(15,5))
    
    # Summarize history for accuracy
    axs[0].plot(range(1,len(model_details.history['acc'])+1),model_details.history['acc'])
    axs[0].plot(range(1,len(model_details.history['val_acc'])+1),model_details.history['val_acc'])
    axs[0].set_title('Model Accuracy')
    axs[0].set_ylabel('Accuracy')
    axs[0].set_xlabel('Epoch')
    axs[0].set_xticks(np.arange(1,len(model_details.history['acc'])+1),len(model_details.history['acc'])/10)
    axs[0].legend(['train', 'val'], loc='best')
    
    # Summarize history for loss
    axs[1].plot(range(1,len(model_details.history['loss'])+1),model_details.history['loss'])
    axs[1].plot(range(1,len(model_details.history['val_loss'])+1),model_details.history['val_loss'])
    axs[1].set_title('Model Loss')
    axs[1].set_ylabel('Loss')
    axs[1].set_xlabel('Epoch')
    axs[1].set_xticks(np.arange(1,len(model_details.history['loss'])+1),len(model_details.history['loss'])/10)
    axs[1].legend(['train', 'val'], loc='best')
    
    # Show the plot
    plt.show()
In [29]:
plot_model(hist)

Step 4: Use a CNN to Classify Dog Breeds

To reduce training time without sacrificing accuracy, we show you how to train a CNN using transfer learning. In the following step, you will get a chance to use transfer learning to train your own CNN.

Obtain Bottleneck Features

In [30]:
bottleneck_features = np.load('bottleneck_features/DogVGG16Data.npz')
train_VGG16 = bottleneck_features['train']
valid_VGG16 = bottleneck_features['valid']
test_VGG16 = bottleneck_features['test']

Model Architecture

The model uses the the pre-trained VGG-16 model as a fixed feature extractor, where the last convolutional output of VGG-16 is fed as input to our model. We only add a global average pooling layer and a fully connected layer, where the latter contains one node for each dog category and is equipped with a softmax.

In [31]:
VGG16_model = Sequential()
VGG16_model.add(GlobalAveragePooling2D(input_shape=train_VGG16.shape[1:]))
VGG16_model.add(Dense(133, activation='softmax'))

VGG16_model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_2 ( (None, 512)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 133)               68229     
=================================================================
Total params: 68,229
Trainable params: 68,229
Non-trainable params: 0
_________________________________________________________________

Compile the Model

In [32]:
VGG16_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])

Train the Model

In [33]:
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.VGG16.hdf5', 
                               verbose=1, save_best_only=True)

hist_VGG16 = VGG16_model.fit(train_VGG16, train_targets, 
          validation_data=(valid_VGG16, valid_targets),
          epochs=20, batch_size=20, callbacks=[checkpointer], verbose=1)
Train on 6680 samples, validate on 835 samples
Epoch 1/20
6660/6680 [============================>.] - ETA: 0s - loss: 12.1836 - acc: 0.1257Epoch 00001: val_loss improved from inf to 10.84747, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 304us/step - loss: 12.1906 - acc: 0.1256 - val_loss: 10.8475 - val_acc: 0.2084
Epoch 2/20
6600/6680 [============================>.] - ETA: 0s - loss: 10.1048 - acc: 0.2777Epoch 00002: val_loss improved from 10.84747 to 10.13946, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 251us/step - loss: 10.1017 - acc: 0.2784 - val_loss: 10.1395 - val_acc: 0.2743
Epoch 3/20
6560/6680 [============================>.] - ETA: 0s - loss: 9.6139 - acc: 0.3396Epoch 00003: val_loss improved from 10.13946 to 9.99965, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 248us/step - loss: 9.6049 - acc: 0.3407 - val_loss: 9.9996 - val_acc: 0.2958
Epoch 4/20
6440/6680 [===========================>..] - ETA: 0s - loss: 9.4644 - acc: 0.3722Epoch 00004: val_loss improved from 9.99965 to 9.92000, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 242us/step - loss: 9.4328 - acc: 0.3734 - val_loss: 9.9200 - val_acc: 0.3066
Epoch 5/20
6580/6680 [============================>.] - ETA: 0s - loss: 9.2557 - acc: 0.3906Epoch 00005: val_loss improved from 9.92000 to 9.71732, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 243us/step - loss: 9.2520 - acc: 0.3909 - val_loss: 9.7173 - val_acc: 0.3281
Epoch 6/20
6620/6680 [============================>.] - ETA: 0s - loss: 9.0652 - acc: 0.4066Epoch 00006: val_loss improved from 9.71732 to 9.55949, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 243us/step - loss: 9.0527 - acc: 0.4072 - val_loss: 9.5595 - val_acc: 0.3353
Epoch 7/20
6580/6680 [============================>.] - ETA: 0s - loss: 8.8894 - acc: 0.4287Epoch 00007: val_loss improved from 9.55949 to 9.45450, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 243us/step - loss: 8.8860 - acc: 0.4289 - val_loss: 9.4545 - val_acc: 0.3413
Epoch 8/20
6660/6680 [============================>.] - ETA: 0s - loss: 8.8281 - acc: 0.4374Epoch 00008: val_loss did not improve
6680/6680 [==============================] - 2s 241us/step - loss: 8.8307 - acc: 0.4373 - val_loss: 9.4759 - val_acc: 0.3353
Epoch 9/20
6460/6680 [============================>.] - ETA: 0s - loss: 8.6989 - acc: 0.4447Epoch 00009: val_loss improved from 9.45450 to 9.32092, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 240us/step - loss: 8.7107 - acc: 0.4440 - val_loss: 9.3209 - val_acc: 0.3581
Epoch 10/20
6480/6680 [============================>.] - ETA: 0s - loss: 8.6596 - acc: 0.4523Epoch 00010: val_loss did not improve
6680/6680 [==============================] - 2s 239us/step - loss: 8.6630 - acc: 0.4518 - val_loss: 9.3654 - val_acc: 0.3437
Epoch 11/20
6460/6680 [============================>.] - ETA: 0s - loss: 8.5656 - acc: 0.4562Epoch 00011: val_loss improved from 9.32092 to 9.16621, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 241us/step - loss: 8.5662 - acc: 0.4558 - val_loss: 9.1662 - val_acc: 0.3593
Epoch 12/20
6460/6680 [============================>.] - ETA: 0s - loss: 8.3987 - acc: 0.4616Epoch 00012: val_loss improved from 9.16621 to 9.06460, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 241us/step - loss: 8.4348 - acc: 0.4596 - val_loss: 9.0646 - val_acc: 0.3617
Epoch 13/20
6460/6680 [============================>.] - ETA: 0s - loss: 8.2360 - acc: 0.4684Epoch 00013: val_loss improved from 9.06460 to 8.90786, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 240us/step - loss: 8.2266 - acc: 0.4686 - val_loss: 8.9079 - val_acc: 0.3689
Epoch 14/20
6440/6680 [===========================>..] - ETA: 0s - loss: 7.9793 - acc: 0.4898Epoch 00014: val_loss improved from 8.90786 to 8.71121, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 241us/step - loss: 8.0076 - acc: 0.4883 - val_loss: 8.7112 - val_acc: 0.3904
Epoch 15/20
6580/6680 [============================>.] - ETA: 0s - loss: 7.9206 - acc: 0.4944Epoch 00015: val_loss improved from 8.71121 to 8.70430, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 245us/step - loss: 7.9217 - acc: 0.4942 - val_loss: 8.7043 - val_acc: 0.3868
Epoch 16/20
6580/6680 [============================>.] - ETA: 0s - loss: 7.7388 - acc: 0.4980Epoch 00016: val_loss improved from 8.70430 to 8.42493, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 245us/step - loss: 7.7257 - acc: 0.4988 - val_loss: 8.4249 - val_acc: 0.4048
Epoch 17/20
6620/6680 [============================>.] - ETA: 0s - loss: 7.5236 - acc: 0.5131Epoch 00017: val_loss improved from 8.42493 to 8.27529, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 243us/step - loss: 7.5074 - acc: 0.5142 - val_loss: 8.2753 - val_acc: 0.4036
Epoch 18/20
6600/6680 [============================>.] - ETA: 0s - loss: 7.4300 - acc: 0.5258Epoch 00018: val_loss did not improve
6680/6680 [==============================] - 2s 243us/step - loss: 7.4134 - acc: 0.5269 - val_loss: 8.3175 - val_acc: 0.4084
Epoch 19/20
6600/6680 [============================>.] - ETA: 0s - loss: 7.3879 - acc: 0.5312Epoch 00019: val_loss did not improve
6680/6680 [==============================] - 2s 243us/step - loss: 7.3889 - acc: 0.5311 - val_loss: 8.3203 - val_acc: 0.4120
Epoch 20/20
6660/6680 [============================>.] - ETA: 0s - loss: 7.3423 - acc: 0.5335Epoch 00020: val_loss improved from 8.27529 to 8.24207, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 2s 244us/step - loss: 7.3397 - acc: 0.5337 - val_loss: 8.2421 - val_acc: 0.4120

Load the Model with the Best Validation Loss

In [34]:
VGG16_model.load_weights('saved_models/weights.best.VGG16.hdf5')

Test the Model

Now, we can use the CNN to test how well it identifies breed within our test dataset of dog images. We print the test accuracy below.

In [35]:
# get index of predicted dog breed for each image in test set
VGG16_predictions = [np.argmax(VGG16_model.predict(np.expand_dims(feature, axis=0))) for feature in test_VGG16]

# report test accuracy
test_accuracy = 100*np.sum(np.array(VGG16_predictions)==np.argmax(test_targets, axis=1))/len(VGG16_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 42.2249%
In [36]:
plot_model(hist_VGG16)

Predict Dog Breed with the Model

In [37]:
from extract_bottleneck_features import *

def VGG16_predict_breed(img_path):
    # extract bottleneck features
    bottleneck_feature = extract_VGG16(path_to_tensor(img_path))
    # obtain predicted vector
    predicted_vector = VGG16_model.predict(bottleneck_feature)
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector)]

print(VGG16_predict_breed('./images/Labrador_retriever_06457.jpg'))
Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5
58892288/58889256 [==============================] - 1s 0us/step
Labrador_retriever

Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)

You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.

In Step 4, we used transfer learning to create a CNN using VGG-16 bottleneck features. In this section, you must use the bottleneck features from a different pre-trained model. To make things easier for you, we have pre-computed the features for all of the networks that are currently available in Keras:

The files are encoded as such:

Dog{network}Data.npz

where {network}, in the above filename, can be one of VGG19, Resnet50, InceptionV3, or Xception. Pick one of the above architectures, download the corresponding bottleneck features, and store the downloaded file in the bottleneck_features/ folder in the repository.

(IMPLEMENTATION) Obtain Bottleneck Features

In the code block below, extract the bottleneck features corresponding to the train, test, and validation sets by running the following:

bottleneck_features = np.load('bottleneck_features/Dog{network}Data.npz')
train_{network} = bottleneck_features['train']
valid_{network} = bottleneck_features['valid']
test_{network} = bottleneck_features['test']
In [39]:
### Obtain bottleneck features from another pre-trained CNN.
import urllib.request
urllib.request.urlretrieve('https://s3-us-west-1.amazonaws.com/udacity-aind/dog-project/DogXceptionData.npz', 
                           'bottleneck_features/DogXceptionData.npz')

bottleneck_features = np.load('bottleneck_features/DogXceptionData.npz')
train_Xception = bottleneck_features['train']
valid_Xception = bottleneck_features['valid']
test_Xception = bottleneck_features['test']

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    <your model's name>.summary()

Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.

Answer:

  1. Use transfer learning to leverage the model which already trained on very large amounts of data.
  2. Use GAP layer(Global Average Pooling) to generate a feature map for each corresponding category and feed directly into the softmax layer. With less parameters it avoids overfitting and speed-up the training time.
In [40]:
### Define your architecture.
from keras.layers import GlobalAveragePooling2D
Xception_model = Sequential()
Xception_model.add(GlobalAveragePooling2D(input_shape=train_Xception.shape[1:]))
Xception_model.add(Dense(133, activation = 'softmax'))
Xception_model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_3 ( (None, 2048)              0         
_________________________________________________________________
dense_3 (Dense)              (None, 133)               272517    
=================================================================
Total params: 272,517
Trainable params: 272,517
Non-trainable params: 0
_________________________________________________________________

(IMPLEMENTATION) Compile the Model

In [41]:
### Compile the model.
Xception_model.compile(loss='categorical_crossentropy', optimizer=Adam(lr=8e-5), metrics=['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [42]:
### Train the model.
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.Xception.hdf5', 
                               verbose=1, save_best_only=True)
hist_Xception = Xception_model.fit(train_Xception, train_targets, 
          validation_data=(valid_Xception, valid_targets),
          epochs=30, batch_size=50, callbacks=[checkpointer], verbose=1, shuffle=True)
Train on 6680 samples, validate on 835 samples
Epoch 1/30
6650/6680 [============================>.] - ETA: 0s - loss: 4.2206 - acc: 0.2114Epoch 00001: val_loss improved from inf to 3.43925, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 365us/step - loss: 4.2179 - acc: 0.2126 - val_loss: 3.4392 - val_acc: 0.5174
Epoch 2/30
6650/6680 [============================>.] - ETA: 0s - loss: 2.7985 - acc: 0.6675Epoch 00002: val_loss improved from 3.43925 to 2.29269, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 290us/step - loss: 2.7963 - acc: 0.6678 - val_loss: 2.2927 - val_acc: 0.7054
Epoch 3/30
6500/6680 [============================>.] - ETA: 0s - loss: 1.8664 - acc: 0.7635Epoch 00003: val_loss improved from 2.29269 to 1.61027, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 288us/step - loss: 1.8579 - acc: 0.7648 - val_loss: 1.6103 - val_acc: 0.7557
Epoch 4/30
6650/6680 [============================>.] - ETA: 0s - loss: 1.3316 - acc: 0.8083Epoch 00004: val_loss improved from 1.61027 to 1.24191, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 290us/step - loss: 1.3311 - acc: 0.8078 - val_loss: 1.2419 - val_acc: 0.7976
Epoch 5/30
6650/6680 [============================>.] - ETA: 0s - loss: 1.0399 - acc: 0.8368Epoch 00005: val_loss improved from 1.24191 to 1.02934, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 291us/step - loss: 1.0384 - acc: 0.8371 - val_loss: 1.0293 - val_acc: 0.8204
Epoch 6/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.8621 - acc: 0.8546Epoch 00006: val_loss improved from 1.02934 to 0.89621, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 288us/step - loss: 0.8616 - acc: 0.8546 - val_loss: 0.8962 - val_acc: 0.8311
Epoch 7/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.7443 - acc: 0.8687Epoch 00007: val_loss improved from 0.89621 to 0.80574, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 291us/step - loss: 0.7445 - acc: 0.8684 - val_loss: 0.8057 - val_acc: 0.8371
Epoch 8/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.6616 - acc: 0.8783Epoch 00008: val_loss improved from 0.80574 to 0.74078, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 289us/step - loss: 0.6608 - acc: 0.8787 - val_loss: 0.7408 - val_acc: 0.8503
Epoch 9/30
6500/6680 [============================>.] - ETA: 0s - loss: 0.5981 - acc: 0.8872Epoch 00009: val_loss improved from 0.74078 to 0.69324, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 288us/step - loss: 0.5977 - acc: 0.8876 - val_loss: 0.6932 - val_acc: 0.8503
Epoch 10/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.5485 - acc: 0.8911Epoch 00010: val_loss improved from 0.69324 to 0.65484, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 290us/step - loss: 0.5481 - acc: 0.8915 - val_loss: 0.6548 - val_acc: 0.8515
Epoch 11/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.5081 - acc: 0.8976Epoch 00011: val_loss improved from 0.65484 to 0.62339, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 290us/step - loss: 0.5081 - acc: 0.8976 - val_loss: 0.6234 - val_acc: 0.8587
Epoch 12/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.4741 - acc: 0.9057Epoch 00012: val_loss improved from 0.62339 to 0.59863, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 287us/step - loss: 0.4739 - acc: 0.9058 - val_loss: 0.5986 - val_acc: 0.8587
Epoch 13/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.4447 - acc: 0.9104Epoch 00013: val_loss improved from 0.59863 to 0.57784, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 290us/step - loss: 0.4453 - acc: 0.9100 - val_loss: 0.5778 - val_acc: 0.8623
Epoch 14/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.4206 - acc: 0.9137Epoch 00014: val_loss improved from 0.57784 to 0.56078, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 290us/step - loss: 0.4201 - acc: 0.9139 - val_loss: 0.5608 - val_acc: 0.8599
Epoch 15/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.3981 - acc: 0.9195Epoch 00015: val_loss improved from 0.56078 to 0.54575, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 289us/step - loss: 0.3980 - acc: 0.9196 - val_loss: 0.5458 - val_acc: 0.8587
Epoch 16/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.3789 - acc: 0.9227Epoch 00016: val_loss improved from 0.54575 to 0.53289, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 290us/step - loss: 0.3785 - acc: 0.9231 - val_loss: 0.5329 - val_acc: 0.8671
Epoch 17/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.3611 - acc: 0.9254Epoch 00017: val_loss improved from 0.53289 to 0.52180, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 289us/step - loss: 0.3605 - acc: 0.9254 - val_loss: 0.5218 - val_acc: 0.8635
Epoch 18/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.3438 - acc: 0.9287Epoch 00018: val_loss improved from 0.52180 to 0.51195, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 288us/step - loss: 0.3440 - acc: 0.9289 - val_loss: 0.5120 - val_acc: 0.8611
Epoch 19/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.3297 - acc: 0.9307Epoch 00019: val_loss improved from 0.51195 to 0.50218, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 291us/step - loss: 0.3291 - acc: 0.9310 - val_loss: 0.5022 - val_acc: 0.8599
Epoch 20/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.3164 - acc: 0.9346Epoch 00020: val_loss improved from 0.50218 to 0.49514, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 302us/step - loss: 0.3159 - acc: 0.9347 - val_loss: 0.4951 - val_acc: 0.8599
Epoch 21/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.3029 - acc: 0.9377Epoch 00021: val_loss improved from 0.49514 to 0.48814, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 299us/step - loss: 0.3031 - acc: 0.9377 - val_loss: 0.4881 - val_acc: 0.8587
Epoch 22/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.2908 - acc: 0.9406Epoch 00022: val_loss improved from 0.48814 to 0.48143, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 308us/step - loss: 0.2911 - acc: 0.9404 - val_loss: 0.4814 - val_acc: 0.8599
Epoch 23/30
6600/6680 [============================>.] - ETA: 0s - loss: 0.2809 - acc: 0.9433Epoch 00023: val_loss improved from 0.48143 to 0.47734, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 303us/step - loss: 0.2803 - acc: 0.9437 - val_loss: 0.4773 - val_acc: 0.8587
Epoch 24/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.2694 - acc: 0.9448Epoch 00024: val_loss improved from 0.47734 to 0.47224, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 290us/step - loss: 0.2697 - acc: 0.9449 - val_loss: 0.4722 - val_acc: 0.8599
Epoch 25/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.2599 - acc: 0.9466Epoch 00025: val_loss improved from 0.47224 to 0.46682, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 290us/step - loss: 0.2598 - acc: 0.9467 - val_loss: 0.4668 - val_acc: 0.8587
Epoch 26/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.2504 - acc: 0.9502Epoch 00026: val_loss improved from 0.46682 to 0.46342, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 291us/step - loss: 0.2505 - acc: 0.9501 - val_loss: 0.4634 - val_acc: 0.8599
Epoch 27/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.2407 - acc: 0.9523Epoch 00027: val_loss improved from 0.46342 to 0.46010, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 289us/step - loss: 0.2418 - acc: 0.9519 - val_loss: 0.4601 - val_acc: 0.8599
Epoch 28/30
6500/6680 [============================>.] - ETA: 0s - loss: 0.2328 - acc: 0.9543Epoch 00028: val_loss improved from 0.46010 to 0.45565, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 288us/step - loss: 0.2337 - acc: 0.9542 - val_loss: 0.4557 - val_acc: 0.8587
Epoch 29/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.2263 - acc: 0.9549Epoch 00029: val_loss improved from 0.45565 to 0.45138, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 288us/step - loss: 0.2259 - acc: 0.9549 - val_loss: 0.4514 - val_acc: 0.8587
Epoch 30/30
6650/6680 [============================>.] - ETA: 0s - loss: 0.2182 - acc: 0.9576Epoch 00030: val_loss improved from 0.45138 to 0.45015, saving model to saved_models/weights.best.Xception.hdf5
6680/6680 [==============================] - 2s 291us/step - loss: 0.2183 - acc: 0.9576 - val_loss: 0.4501 - val_acc: 0.8611

(IMPLEMENTATION) Load the Model with the Best Validation Loss

In [43]:
### Load the model weights with the best validation loss.
Xception_model.load_weights('saved_models/weights.best.Xception.hdf5')

(IMPLEMENTATION) Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 60%.

In [44]:
### Calculate classification accuracy on the test dataset.
Xception_predictions = [np.argmax(Xception_model.predict(np.expand_dims(feature, axis=0))) for feature in test_Xception]

test_accuracy = 100*np.sum(np.array(Xception_predictions)==np.argmax(test_targets, axis=1))/len(Xception_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 85.5263%
In [45]:
plot_model(hist_Xception)

(IMPLEMENTATION) Predict Dog Breed with the Model

Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan_hound, etc) that is predicted by your model.

Similar to the analogous function in Step 4, your function should have three steps:

  1. Extract the bottleneck features corresponding to the chosen CNN model.
  2. Supply the bottleneck features as input to the model to return the predicted vector. Note that the argmax of this prediction vector gives the index of the predicted dog breed.
  3. Use the dog_names array defined in Step 0 of this notebook to return the corresponding breed.

The functions to extract the bottleneck features can be found in extract_bottleneck_features.py, and they have been imported in an earlier code cell. To obtain the bottleneck features corresponding to your chosen CNN architecture, you need to use the function

extract_{network}

where {network}, in the above filename, should be one of VGG19, Resnet50, InceptionV3, or Xception.

In [46]:
### Write a function that takes a path to an image as input
### and returns the dog breed that is predicted by the model.

def Xception_predict_breed(img_path):
    bottleneck_feature = extract_Xception(path_to_tensor(img_path))
    predicted_vector = Xception_model.predict(bottleneck_feature)
    return dog_names[np.argmax(predicted_vector)]

print(Xception_predict_breed('./images/Labrador_retriever_06457.jpg'))
Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.4/xception_weights_tf_dim_ordering_tf_kernels_notop.h5
83689472/83683744 [==============================] - 1s 0us/step
Labrador_retriever

Step 6: Write your Algorithm

Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,

  • if a dog is detected in the image, return the predicted breed.
  • if a human is detected in the image, return the resembling dog breed.
  • if neither is detected in the image, provide output that indicates an error.

You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and dog_detector functions developed above. You are required to use your CNN from Step 5 to predict dog breed.

Some sample output for our algorithm is provided below, but feel free to design your own user experience!

Sample Human Output

(IMPLEMENTATION) Write your Algorithm

In [54]:
### Write your algorithm.
### Feel free to use as many code cells as needed.

from os import walk

def plot_results(file_paths):
    rows = len(file_paths)
    k = 0
    for i in file_paths:
        dog_breed = Xception_predict_breed(i)
        for root, dirs, files in walk('dogImages/train/'):
            if dog_breed in root:
                family = root + '/' + files[0]
                break
        fig = plt.figure(figsize=(30,30))
        k += 1
        fig.add_subplot(rows, 3, k)
        plt.imshow(np.expand_dims(ndimage.imread(i), 0)[0])
        if face_detector2(i):
            plt.title('Hi Human! Your dog breed is {}'.format(dog_breed), fontdict={'fontsize':20})
            k += 1
            fig.add_subplot(rows, 3, k)
            plt.imshow(np.expand_dims(ndimage.imread('test_images/arrow.png'), 0)[0])  
            plt.axis('off')
            k += 1
            fig.add_subplot(rows, 3, k)
            plt.imshow(np.expand_dims(ndimage.imread(family), 0)[0])
            plt.title('This is your family!', fontdict={'fontsize':20})
        elif dog_detector(i):
            plt.title('Hi Dog! I guess your breed is {}'.format(dog_breed), fontdict={'fontsize':20})
            k += 1
            fig.add_subplot(rows, 3, k)
            plt.imshow(np.expand_dims(ndimage.imread('test_images/arrow.png'), 0)[0])
            plt.axis('off')
            k += 1
            fig.add_subplot(rows, 3, k)
            plt.imshow(np.expand_dims(ndimage.imread(family), 0)[0])
            plt.title('This is your family!', fontdict={'fontsize':20})
        else:
            plt.title('You are neither human nor dog!', fontdict={'fontsize':20})
    
    plt.show() 

Step 7: Test Your Algorithm

In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?

(IMPLEMENTATION) Test Your Algorithm on Sample Images!

Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.

Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.

Answer:

The output result is better than expected. It successfully detected dogs and gave reasonable dog breeds. It also gave the woman an interesting related dog-breed. The reason why it cannot detect a man with a laid-back posture is because the large angle of face tilt. In order to further improve the algorithm, we can

  1. Use different pre-trained models in keras for transfer learning, e.g. DenseNet, NASNet.
  2. Use image augmentation for pre-training.
  3. Collect more labeled data.
In [48]:
!mkdir test_images
In [55]:
## Execute your algorithm from Step 6 on
## at least 6 images on your computer.
## Feel free to use as many code cells as needed.

import urllib.request
urllib.request.urlretrieve('https://cdn.pixabay.com/photo/2017/04/06/09/16/arrow-2207748_960_720.png','test_images/arrow.png')
urllib.request.urlretrieve('https://cdn.pixabay.com/photo/2018/04/27/16/49/dog-3355192_960_720.jpg','test_images/Boxer.jpg')
urllib.request.urlretrieve('https://cdn.pixabay.com/photo/2018/02/18/15/00/dog-3162483_960_720.jpg','test_images/Labrador_retriever.jpg')
urllib.request.urlretrieve('https://cdn.pixabay.com/photo/2016/06/11/18/00/dog-1450447_960_720.jpg','test_images/Pomeranian.jpg')
urllib.request.urlretrieve('https://cdn.pixabay.com/photo/2018/03/27/21/50/nature-3267539_960_720.jpg','test_images/Golden_retriever.jpg')
urllib.request.urlretrieve('https://cdn.pixabay.com/photo/2018/02/21/15/06/woman-3170568_960_720.jpg','test_images/woman.jpg')
urllib.request.urlretrieve('https://cdn.pixabay.com/photo/2017/06/28/04/29/adult-2449725_960_720.jpg','test_images/man.jpg')

file_paths = ['test_images/Boxer.jpg',
              'test_images/Labrador_retriever.jpg',
              'test_images/Pomeranian.jpg',
              'test_images/Golden_retriever.jpg', 
              'test_images/woman.jpg',
              'test_images/man.jpg']
plot_results(file_paths)